Sustainable experimental design with ai

Abstract

A method for controlling one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s), the method comprising: providing a target parameter range indicative of one or more target parameter value(s) associated with a target parameter characterizing the one or more target chemical reaction(s) determining one or more suggestion parameter value(s) associated with a tuneable parameter by an operation data-driven model from the one or more target parameter value(s), wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, and wherein the operation data-driven model is determined according to any one of the method claims, providing the one or more suggestion parameter value(s) for controlling the one or more target chemical reaction(s) according to the one or more suggestion parameter value(s).

Description

[0001] SUSTAINABLE EXPERIMENTAL DESIGN WITH Al

[0002] TECHNICAL FIELD

[0003] The invention relates to sustainable design of experiments and a method for controlling one or more target chemical reaction(s), a method for determining an operation data-driven model configured to determine parameter values for controlling one or more target chemical reaction(s), use of an extraction data-driven model, use of an operation data-driven model, apparatus for controlling one or more target chemical reaction(s).

[0004] TECHNICAL BACKGROUND

[0005] Chemicals are the basis for materials used and produced in industry. Therefore, chemical products need to be tailored towards their intended use by developing efficient production procedures. This requires a lot of time and resources. Hence, it is desired to shorten the development time for chemical production procedures and increase the resource-efficiency with respect to experimenting.

[0006] SUMMARY

[0007] In an aspect, this disclosure relates to a, in particular computer-implemented, method for controlling one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s), the method comprising: providing a target parameter range indicative of one or more target parameter value(s) associated with a target parameter characterizing the one or more target chemical reaction(s) determining one or more suggestion parameter value(s) associated with a tuneable parameter by an operation data-driven model from the one or more target parameter value(s), wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, and wherein the operation data-driven model is determined by receiving experimental data comprising a plurality of parameter values associated with a plurality of parameters characterizing the one or more target chemical reaction(s), wherein the plurality of parameters comprises at least one target parameter and at least one tuneable parameter, wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, and providing model determining task instructions including the experimental data for providing the operation data-driven model to an extraction data-driven model, wherein the extraction data-driven model is configured to follow task instructions, and wherein the operation data-driven model is configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data, providing the one or more suggestion parameter value(s) for controlling the one or more target chemical reaction(s) according to the one or more suggestion parameter value(s).

[0008] In another aspect, it relates to a method for determining an operation data-driven model configured to determine parameter values for controlling one or more target chemical reaction(s), the method comprising: receiving experimental data comprising a plurality of parameter values associated with a plurality of parameters characterizing the one or more target chemical reaction(s), wherein the plurality of parameters comprises at least one target parameter and at least one tuneable parameter, wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, providing model determining task instructions including the experimental data for providing the operation data-driven model to an extraction data-driven model, wherein the extraction data-driven model is configured to follow task instructions, and wherein the operation data-driven model is configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data, providing the operation data-driven model, in particular for providing parameter values for controlling one or more target chemical reaction(s).

[0009] In another aspect, it relates to use of an extraction data-driven model, in particular as described herein, for extracting at least one set of parameter values associated with at least one target parameter and at least one tuneable parameter for training an operation data-driven model, in particular as described herein, based on the at least one set of parameter values.

[0010] In another aspect, it relates to use of an operation data-driven model trained as described herein for determining parameter values for controlling one or more target chemical reaction(s).

[0011] In another aspect, it relates to a non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to perform any one of the methods as presented herein.

[0012] In another aspect, it relates to use of experimental data including at least one set of parameter values associated with at least one target parameter and at least one tuneable parameter extracted by an extraction data-driven model for training an operation data-driven model according to any one of the methods as described herein. In another aspect, it relates to an apparatus for controlling one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s), the apparatus comprising: a processor; and a memory storing instructions that, when executed by the processor, configure the apparatus to perform the steps of any one of the methods as described herein.

[0013] EMBODIMENTS

[0014] In the following, terminology as used herein and / or the technical field of the present disclosure will be outlined by ways of definitions and / or examples. Where examples are given, it is to be understood that the present disclosure is not limited to said examples.

[0015] These and other objects, which become apparent upon reading the following description, are solved by the subject matters of the independent claims. The dependent claims refer to embodiments of the invention.

[0016] Chemical products are produced on diverse scales to arrive at highly efficient large-scale production procedures. This requires a intensive testing of production procedures starting on small scales developing through medium scale batch sizes to large-scale chemical plants processing tons of chemical products. In each phase, the production procedure needs to be adapted to product the chemical product having the target properties, i.e. , producing the target chemical product associated with the target parameter values. Followingly, testing chemical production procedures involves a plurality of testing series for evaluation parameters of a chemical reaction. This requires a lot of materials, energy and water while costing time of production facilities and workers. Furthermore, chemical waste is produced that needs to be treated expensively. Currently, developing production routines involves educated guesses by experts as depicted in FIG. 2A. This is prone to subjective suggestions by the experts, i.e. unnecessarily limited towards personal experience and / or comprise of the respective expert. Thus, it is desired to improve developing chemical production procedures by reducing the number of experiments carried out.

[0017] Providing model determining task instructions including the experimental data for providing an operation data- driven model to an extraction data-driven model allows to build a suggestion model for suggesting parameter values of chemical production procedures from experimental data. The experimental data may include an extensive library of already conducted experiments. Said experimental data may suggest a plurality of parameter values for chemical production procedures. The extraction data-driven model is capable of understanding the content of the experimental data and derive trends, i.e. interpolate and / or extrapolate from the provided experimental data. Thus, the extraction data-driven model can extract more information than the explicitly mentioned parameter values, in particular on large scale. Hence, the extraction data-driven model may extract and combine knowledge from different sources in a small time frame. As a consequence, the extraction data-driven model may make use of historical experiments carried out and obtain even more information from the experimental data than included, e.g. via qualitative and / or quantitative relations. This improved knowledge extraction results in building reliable operation data-driven models. Said operation data- driven models guide researches in finding the optimal production procedures. This guidance reduces the number of experimentation cycles carried out by a factor of 3 or 4 in comparison to random guessing as it can be seen in FIG. 2A, FIG. 2B and FIG. 7. Ultimately, this allows to conduct more testing with the same number of chemical production facilities, i.e. laboratories and chemical plants. Additionally or alternatively, less resources are consumed when developing production procedures. Furthermore, this procedure allows to react faster upon the increasing the demand of chemical products in an increasingly complex and fast developing world.

[0018] In an embodiment, providing input data to the data-driven model may comprise mapping the input data to a numerical representation of the input data. The numerical representation of the input data may comprise a tensor associated with the input data and / or obtained from the input data. In particular, the numerical representation of the input data may be indicative of two or more elements of the input data and a relation between the two or more elements of the input data. Preferably, providing input data to the data-driven model may comprise at least one of identifying two or more elements of the input data, mapping the two or more elements of the input data to a numerical representation of the two or more elements, mapping the numerical representation of the two or more elements to a numerical representation of a predefined size related to the numerical representation of the two or more elements, mapping the numerical representation of the predefined size related to the numerical representation of the two or more elements to a numerical representation of the two or more elements and a relation between the two or more elements or a combination thereof. Providing the extraction task instruction includes mapping the extraction task instruction to a numerical representation of the extraction task instruction. The extraction data-driven model may be configured to map the numerical representation of the extraction task instruction to the at least one set of parameter values, in particular a numerical representation of the at least one set of parameter values. Any one of the methods may further comprise mapping the numerical representation of the at least one set of parameter values to the at least one set of parameter values. The numerical representation may be processed and / or may be processable by the extraction data-driven model. The numerical representation may be a structured representation.

[0019] In an embodiment, processing the input data and / or generating the output data from the input data may comprise processing the numerical representation of the input data, in particular the numerical representation of the two or more elements and the relation between the two or more elements. Processing the numerical representation of the two or more elements and the relation between the two or more elements may comprise mapping the numerical representation of the two or more elements, and optionally the relation between the two or more elements to a numerical representation of the output data. The numerical representation of the output data may be mapped to the output data, in particular based on a relation between the numerical representation of data and the data. In particular, the data may be of a data type according to the input data. Hence, the output data may be of the data type according to the input data, e.g. of the same data type as the input data and / or of the data type specified by the input data. Preferably, processing the numerical representation of the two or more elements and the relation between the two or more elements may comprise at least one of generating two or more numerical representations of the two or more elements and the relation between the two or more elements from the numerical representation of the two or more elements and the relation between the two or more elements, modifying the two or more numerical representation of the two or more elements and the relation between the two or more elements by applying a filter to the two or more numerical representations of the two or more elements and the relation between the two or more elements, concatenating the two or more numerical representations of the two or more elements and the relation between the two or more elements, mapping the concatenated numerical representation of the two or more elements and the relation between the two or more elements to a numerical representation of the output data or a combination thereof.

[0020] In an embodiment, task instruction may include experimental data, proposed experimental data and / or one or more instructions for triggering the extraction data-driven model to provide the operation data-driven model and / or to extract the at least one set of parameter values associated with the at least one target parameter and the at least one tuneable parameter from the experimental data. The at least one set of parameter values may comprise at least one parameter value associated with the at least one target parameter and at least one parameter value associated with the at least one tuneable parameter. The task instruction may be input data to the extraction data-driven model. Processing of the task instruction by the extraction data-driven model may trigger the extraction data-driven model to follow the task instructed by the task instruction, e.g. Extract the at least one set of parameter values and / or provide the operation data-driven model.

[0021] In an embodiment, the experimental data may be recorded in relation to the plurality of chemical reactions. Hence, the experimental data may comprise sensor data obtained in relation to the plurality of chemical reactions and / or obtained by processing the sensor data via one or more mathematical operation(s). For example, the sensor data may be converted into another numerical schema or a measure in relation to the sensor data may be obtained based on a mathematical formular applied to the sensor data.

[0022] In an embodiment, data-driven model may refer to a model suitable for describing one or more relations between input data and output data, in particular non-linear relations between input data and output data. Input data may refer to data to be provided to the data-driven model and / or to data being received by the data- driven model. Output data may be data to be received from the data-driven model and / or to be provided by the data-driven model. Hence, the data-driven model may determine the output data based on transforming the input data via one or more non-linear relations. The data-driven model may obtain the relation between the input data and the output data during the training of the data-driven model. In an embodiment, adapting the parameter values associated with the tuneable parameter may change the parameter values associated with the target parameter, in particular increase or decrease the parameter values associated with the target parameter.

[0023] In an embodiment, the extraction task instruction data may include string data and / or a sequence of one or more elements. The one or more element(s) may comprise at least a part of a word, a number, a symbol or the like. Thereby, the extraction task instruction may be user-interpretable. In chemical production, chemical plants need to be operated by longly experienced operators. Therefore, it is desired to support operators of chemical plants in adjusting chemical reaction procedures. This allows for an easy instructing of the extraction data-driven model by human users as the human is supported in interacting with the models. Further, output from the model becomes interpretable. Thus, users are enabled to identify errors and intervene. Ultimately, this improves controlling of chemical reactions.

[0024] In an embodiment, providing model determining task instructions to the extraction data-driven model may comprise: providing extraction task instructions including the experimental data to an extraction data-driven model for extracting at least one set of parameter values associated with the at least one target parameter and the at least one tuneable parameter from the experimental data. The extraction data-driven model may be configured to follow task instructions, and training an operation data-driven model to relate parameter values associated with the at least one tuneable parameter to the at least one target parameter according to the at least one extracted set of parameter values. Training the operation data-driven model may include a predefined number of parameters of the operation data-driven model and updating the predefined number of parameters to reduce a deviation between parameter values associated with the at least one tuneable parameter or the at least one target parameter determined by the operation data-driven model and the parameter values associated with the at least one tuneable parameter or the at least one target parameter according to the experimental data. The deviation may be reduced until the deviation may be within a predefined deviation range. The operation data-driven model may comprise one or more mathematical relation(s), in particular function(s) relating the parameter values associated with the at least one tuneable parameter and the parameter values associated with the at least one target parameter. The extraction task instructions may trigger the extraction data-driven model to extract at least one set of parameter values associated with the at least one target parameter and the at least one tuneable parameter from the experimental data. The extraction data-driven model may select the at least one set of parameter values associated with the at least one target parameter and the at least one tuneable parameter. One set of parameter values associated with the at least one target parameter and the at least one tuneable parameter may comprise at least two parameter values. The at least one parameter values may be associated with the at least one target parameter and at least one parameter values may be associated with the at least one tuneable parameter. By extracting the at least one set from the experimental data, already available information may be used to build an operation data-driven model based on the at least one extracted set. Thereby, less or no further experiments may be required to determine control parameters for target chemical reactions. Ultimately, significant resources for developing target chemical reactions that allow to obtain target chemical products with target properties can be saved.

[0025] In an embodiment, determining task instructions to the extraction data-driven model may comprise providing model determining task instructions including the experimental data for generating an operation data-driven model to an extraction data-driven model. The operation data-driven model may be provided by the extraction data-driven model. Optionally, any one of the methods may further comprise providing the operation data- driven model for determining one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s). At least a part of the parameter value(s) may be used for controlling the one or more target chemical reaction(s) according to one or more parameter value(s). The experimental data may be indicative of and / or may include one or more relation(s) between the parameter values associated with the at least one target parameter and the at least one tuneable parameter. The operation data-driven model may comprise the one or more relation(s) between the parameter values associated with the at least one target parameter and the at least one tuneable parameter. Hence, the extraction data-driven model may extract the one or more relation(s) between the parameter values associated with the at least one target parameter and the at least one tuneable parameter from the experimental data. Experimental data are typically indicative and / or experimental data are based on scientific relations. This allows to extract relations already present in the experimental data or already obtained in historical experiments by the extraction data-driven model. The extraction data-driven model may be trained based on one or more physical, chemical and / or biological equation(s). Hence, the extraction data-driven model may have obtained relations between parameters during the training. By extracting the operation data- driven model from the experimental data, already available information may be used to build an operation data-driven model. Thereby, less or no further experiments may be required to determine control parameters of target chemical reactions. Ultimately, significant resources for developing target chemical reactions that allow to obtain target chemical products with target properties can be saved.

[0026] In an embodiment, the extraction data-driven model may be trained based on historical extraction task instructions and corresponding extracted sets of parameter values associated with two or more parameters. The extraction data-driven model may be trained and / or may be configured to extract parameter values from experimental data according to extraction task instructions. The extraction data-driven model may be a finetuned data-driven model.

[0027] In an embodiment, the extraction data-driven model may be a pretrained data-driven model. The pretrained data-driven model may be parametrized and / or trained based on data with a plurality of contexts and / or unstructured data, in particular text data and optionally numerical data such as tabular data or image data. The pretrained data-driven models may be configured to perform a plurality of task and / to process data of a plurality of contexts. The pretrained data-driven models may be configured to perform the task according to the provided task instruction. Hence, the pretrained data-driven model may be configured to be provided with a plurality of different task instructions and / or provide a plurality of different types of output data upon receiving different task instructions. Thereby, already available models can be directly used. Hence, no resources for further training are necessary and on premise-models can be used. As a consequence, pretrained data-driven models are useful on small to medium-sized scales.

[0028] In an embodiment, the finetuned data-driven model may be obtained by training a pretrained data-driven model. The finetuned data-driven models may be trained additionally on a training data set comprising a plurality of extraction task instructions and corresponding extracted sets of parameter values. Finetuning a model for a specific application allows to improve the performance, i.e. accuracy and precision, of the model. Hence, finetuning the extraction data-driven model improves the reliability of extracting parameter values, in particular according to predefined criteria. Followi ngly , less extraction instructions are needed to trigger extracting by the extraction data-driven model allowing for less data processing in order to extract parameter values. This is advantageous where large amounts of parameter values need to be extracted. Therefore, finetuning decreases the resources for extracting parameter values on large scales.

[0029] In an embodiment, the extraction data-driven model may be a generative data-driven model. In an embodiment, the extraction data-driven model may be a language model, in particular a large language model.

[0030] In an embodiment, determining the operation data-driven model may comprise training the data-driven model and / or providing, in particular generating the operation data-driven model by the extraction data-driven model.

[0031] In an embodiment, a task instruction may refer to a prompt.

[0032] Any one of the methods may further comprise controlling the one or more target chemical reaction(s) according to the one or more suggestion parameter value and / or providing the one or more target chemical reaction(s) according to the one or more suggestion parameter value to a control engine associated with controlling the one or more target chemical reaction(s).

[0033] In an embodiment, any one of the methods may further comprise determining proposed experimental data according to a distribution of the parameter values associated with the at least one target parameter and the at least one tuneable parameter within the experimental data, and / or determining the proposed experimental data according to a relation between the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter. The proposed experimental data may comprise at least one proposed set of parameter values associated with at least one target parameter and at least one tuneable parameter. The extraction task instructions may further include the proposed experimental data. The experimental data may be indicative of and / or may comprise the distribution of the parameter values associated with the at least one target parameter and the at least one tuneable parameter. Determining the proposed experimental data may comprise determining at least two proposed parameter values associated with the at least one target parameter and the at least one tuneable parameter. In particular, at least one of the two proposed parameter values may be associated with the at least one target parameter and / or the at least one tuneable parameter. Determining the proposed experimental data may be based on a distribution function of parameter values associated with the at least one target parameter and / or the at least one tuneable parameter, in particular a cumulative distribution function relating parameter values associated with the at least one target parameter and / or the at least one tuneable parameter. The distribution function may be indicative of a distribution of parameter values, in particular a likelihood of an appearance of parameter values associated with the at least one target parameter and / or the at least one tuneable parameter. Determining the distribution function may comprise dividing the distribution function into a number of parts equal to a number of corresponding parameter values and / or a number of sets of parameter values and / or assigning at least one parameter value per part. The relation between the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter may be a mathematical function relating the parameter values associated with the at least one target parameter to the parameter values associated with the at least one tuneable parameter or vice versa. The relation may be obtained from a database and / or may be extracted by one or more data-driven model(s) such as the extraction data-driven model. For this purpose, a data extraction task instruction for extracting the proposed experimental data from the experimental data. The one or more data-driven model(s) may be configured to follow a plurality of different task instructions. By doing so, parameter values similar to the parameter values comprised by the experimental data may be obtained. The so-determined parameter values may be indirectly included in the experimental data via relations between the target parameter and the tuneable parameter. This allows to obtain more datapoints for building the operation data-driven model than directly obtained from the experimental data. Consequently, model performance of the operation data-driven model is improved leading to more reliable and diverse parameter values provided by the operation data- driven model. Ultimately, this reduces the number of experiments to be run to arrive at the one or more target chemical reaction(s).

[0034] In an embodiment, the experimental data may comprise proposed experimental data including at least one proposed set of parameter values associated with at least one target parameter and at least one tuneable parameter. The proposed experimental data may be determined according to a distribution of the parameter values associated with the at least one target parameter and the at least one tuneable parameter within historical experimental data, and / or determining the proposed experimental data according to a relation between the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter. In an embodiment, the relation and / or the distribution may be obtainable and / or may be obtained from the experimental data. The experimental data may be indicative of the relation and / or the distribution.

[0035] In an embodiment, any one of the methods may further comprise providing an applicable parameter range based on an inventory of a target chemical production facility for obtaining the target chemical product and / or one or more setting(s) related to the target chemical production facility and selecting at least a part of the experimental data according to the applicable parameter range. The experimental data provided to the extraction data-driven model may correspond to the selected part of the experimental data. Additionally or alternatively, the part of the experimental data may be selected according to the one or more target chemical reaction(s). The experimental data may be indicative of the one or more target chemical reaction(s), e.g., by comprising an indication of the one or more target chemical reaction(s). Additionally or alternatively, the indication of the one or more target chemical reaction(s) may be provided and / or received. The part of the experimental data may be selected according to the indication of the one or more target chemical reaction(s). Selecting the part of the experimental data according to the one or more target chemical reaction(s) may comprise matching the chemical reactions associated with the experimental data and the one or more target chemical reaction(s) e.g. according to the indication of the one or more target chemical reaction(s). Additionally or alternatively, the one or more parts of the experimental data may be mapped to a numerical representation of the one or more parts of the experimental data and at least one part of the experimental data may be selected by determining a distance between the numerical representation of the one or more parts of the experimental data and the indication of the one or more target chemical reaction(s), in particular a numerical representation of the one or more target chemical reaction(s). The applicable parameter range may comprise at least the subset of available and / or used parameter values. This allows to preselect meaningful parameter values with respect to the tuneable parameters. For example, parameter values within a predefined range may allow to operate chemical production facilities safely and / or in a controlled manner. Hence, reducing the search space of the tuneable parameter values enables to preselect workable parameter values and make use of promising synthesis routes already developed. This saves time and resources for experimentation while increasing the accuracy of the operation data-driven model built upon the preselected parameter values. Ultimately, this further reduces the number of experimental cycles needed in the development of chemical synthesis routes.

[0036] In an embodiment, any one of the methods may further comprise determining an applicable parameter range from the experimental data by the extraction data-driven model. The applicable parameter range may be indicative of a range of tuneable parameter values associated with at least one tuneable parameter. The experimental data provided to the extraction data-driven model may correspond to the selected part of the experimental data. The extraction task instructions may further comprise one or more instruction(s) for triggering the extraction data-driven model to provide the applicable parameter range. The experimental data may be indicative of the applicable parameter range. The experimental data may comprise a subset of available and / or used parameter values, i.e. the tuneable parameter values, associated with the at least one tuneable parameter. The applicable parameter range may comprise at least the subset of available and / or used parameter values. This allows to preselect meaningful parameter values with respect to the tuneable parameters. For example, parameter values within a predefined range may allow to operate chemical production facilities safely and / or in a controlled manner. Hence, reducing the search space of the tuneable parameter values enables to preselect workable parameter values and make use of promising synthesis routes already developed. This saves time and resources for experimentation while increasing the accuracy of the operation data-driven model built upon the preselected parameter values. Ultimately, this further reduces the number of experimental cycles needed in the development of chemical synthesis routes.

[0037] In an embodiment, any one of the methods may further comprise providing verification experimental data indicative of at least one verification parameter value associated with the at least one target parameter. The verification experimental data may be obtained during the one or more target chemical reaction(s) controlled according to the one or more suggestion parameter value(s) and providing a trigger for adapting the operation data-driven model in response to determining that the operation data-driven model may be updated upon the verification experimental data. Any one of the methods may further comprise determining if the operation data- driven model may be updated based on the verification experimental data. Determining if the operation data- driven model may be updated based on the verification experimental data may include determining a deviation between the one or more parameter values associated with the target parameter range and at least one verification parameter value. Determining the deviation may comprise determining a deviation score, in particular a numerical value. For example, the deviation score may allow to rank deviations between parameter values associated with the target parameter range and verification parameter values on a scale. Additionally or alternatively, the deviation score may be a measure for the deviation related to a standard deviation associated with the proposed experimental data and / or the suggestion parameter values. In particular, determining the deviation may comprise determining a distance between the one or more parameter value(s) associated with the target parameter range and at least one verification parameter value. A measure for the distance may be related to a standard deviation associated with the proposed experimental data and / or the suggestion parameter values. Optionally, the parameter value associated with the target parameter may be determined from the verification experimental data.

[0038] In an embodiment, any one of the methods may further comprise providing verification experimental data for verifying the operation data-driven model. Verifying the operation data-driven model may comprise providing verification experimental data indicative of at least one verification parameter value associated with the at least one target parameter.

[0039] In an embodiment, the operation data-driven model may comprise a combination of two or more data-driven model(s) configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data. The combination of the two or more data-driven models may obtain parameter values by relating the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter via the two or more data-driven models, preferably by relating the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter per data-driven model amd weighting the parameter values obtained per data-driven model. Preferably, the parameter values obtained per data-driven model may be weighted according to a confidence score associated with the two or more data-driven models. The confidence score may be indicative of an accuracy associated with the two or more data-driven models. Preferably, two or more confidence scores may be associated per data-driven model, in particular per range of parameter values and per data-driven model. The two or more confidence scores may be associated with different ranges of parameter values per data-driven model. The operation data-driven model may relate the two or more data-driven models via two or more weighting factors obtained based on the confidence scores associated with the data-driven models, and optionally associated with a subset of the parameter values. For example, the first data-driven model may be associated with a first range of parameter values and a second range of parameter values. The first data-driven model may be associated with a first accuracy within the first range of parameter values and a second accuracy different from the first accuracy within the second range of parameter values. Where the confidence score associated with the first range of parameter values and the first data-driven model may be within a predefined confidence range, the weighting factor for weighting the influence of the two or more data-driven models may be higher in the first range of parameter values with respect to the first data-driven model than the weighting factor in the first range of parameter values with respect to the second data-driven model. In an embodiment, the two or more data-driven models may comprise a sub model and the extraction data-driven model. The sub model may be trained and / or determined analogous to determining and / or training the operation data-driven model. Additionally or alternatively, the operation data-driven model may comprise a combined function. The combined function may be obtained by relating two or more function(s) associated with the two or more data- driven models via two or more weighting factors associated with the two or more data-driven models. At least one of the two or more data-driven models may be configured according to the experimental data, in particular the verification experimental data and independent of proposed experimental data. By doing so, insights into the relation between the target parameter and the tuneable parameter can be obtained from different models, e.g. with a different focus and / or with different advantages and disadvantages. Thereby, the best from both models can be used to provide the suggestion parameter values with a higher accuracy and / or requiring less verifications via verification experimental data. At least one of the two or more data-driven models may be trained and / or configured based on historical experimental data, in particular historical experimental data obtained from sensor data, e.g. by transforming sensor data via one or more mathematical operation(s). Additionally or alternatively, at least one of the two or more data-driven models may be trained and / or configured based on historical proposed experimental data.

[0040] In an embodiment, the experimental data may be indicative of the one or more target chemical reaction(s) and / or any one of the methods may further comprise providing the indication of the one or more target chemical reaction(s). A part of the experimental data may be selected according to the indication of the one or more target chemical reaction(s). The experimental data provided to the extraction data-driven model may correspond to the selected part of the experimental data. By doing so, the extraction data-driven model is provided with a preselection of the experimental data, i.e. the relevant experimental data. This allows to focus the attention of the data-driven model to the relevant part of the experimental data and hence, improves extracting of the parameter values by the extraction data-driven model. Ultimately, this improves building the operation data-driven model and hence improve the suggestion parameter values. Thereby, less experimentation cycles are needed to arrive at the target specifications.

[0041] In an embodiment, the target parameter may include and / or may be a tuneable parameter. Preferably, the target parameter may be a different tuneable parameter than the tuneable parameter associated with the one or more suggestion parameter value(s).

[0042] In an embodiment, providing the model determining task instructions may trigger the extraction data-driven model to provide an operation data-driven model and / or provide training data for training the operation data- driven model. Providing the model determining task instructions may result in training the operation data- driven model according to at least a part of the experimental data and / or data extracted by the extraction data- driven model from the experimental data and / or providing one or more model parameter(s) for relating the parameter values associated with the at least one tuneable parameter and the at least one target parameter.

[0043] BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0044] In the following, the present disclosure is further described with reference to the enclosed figures. The same reference numbers in the drawings and this disclosure are intended to refer to the same or like elements, components, and / or parts.

[0045] FIG. 1 illustrates an embodiment of producing chemical products 116 by one or more chemical production facilities 102.

[0046] FIG. 2A illustrates an embodiment of determining one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0047] FIG. 2B illustrates an embodiment of determining one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0048] FIG. 3 illustrates an embodiment of a method for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0049] FIG. 4 illustrates an embodiment of a method for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s). FIG. 5 illustrates an operating system of a chemical production entity for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0050] FIG. 6 illustrates an embodiment of a user interface for obtaining a target chemical product.

[0051] FIG. 7 illustrates evolution of a parameter values of a target parameter in relation to a number of iterations including conducting experiments.

[0052] FIG. 8 illustrates an embodiment of training an embedding layer.

[0053] DETAILED DESCRIPTION

[0054] The following embodiments are mere examples for implementing the method, the system or application device disclosed herein and shall not be considered limiting.

[0055] FIG. 1 illustrates an embodiment of producing chemical products 116 by one or more chemical production facilities 102.

[0056] Chemical products may be produced by the one or more chemical production facilities 102. The chemical products 116 produced by the one or more chemical production facilities 102 may be associated with one or more target parameter value 112. The chemical products 116 may be processed by one or more product processing facilities 114. For processing the chemical products 116 by the one or more product processing facilities 114 the chemical products 116 may be desired to have to one or more target parameter value 112. The one or more target parameter value 112 may be necessary for producing end products with a desired quality. Said end products may be versatile and may range from parts of cars to packaging. End products with a lower quality may become waste immediately as such end products may not be suited for the intended application or at least faster than the end products with the desired quality due to faster degradation. Hence, ensuring sufficient quality of the chemical products produced by the one or more chemical production facilities 102 is important to reduce resource waste. Thereby, the waste of chemical production is further lowered, i.e., the efficiency of producing chemical products 116 is increased, while the production of the chemical products by the one or more chemical production facilities 102 may be tailored to the desired properties of the end products. The challenge in chemical production is to select the target reaction conditions for producing target chemical products associated with target property values. Said target reaction conditions may be obtained as described in the context of FIG. 2A and FIG. 2B. FIG. 2A illustrates an embodiment of determining one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0057] For example, the chemical reaction such as a SN2 reaction of two educts for forming two products. Parameters characterizing one or more target chemical reaction(s) may include reaction condition(s) and / or one or more chemical(s) participating in the one or more target chemical reaction(s). For example, the one or more chemical(s) may include at least one of one or more educt(s), one or more product(s), one or more catalyst(s), one or more solvent(s), one or more substrate(s) or a combination thereof. The one or more target chemical reaction(s) may be conducted to obtain and / or produce one or more target chemical product(s). Target parameter values associated with target parameters may characterize targets in relation to the one or more target chemical product(s). Target parameter values may be desired for processing chemical products towards end products as described in the context of FIG. 1.

[0058] Typically, target parameter values are obtained by a plurality of iterations including gathering historical experimental data 202 by a chemical expert 204, defining a target parameter range 218 by the chemical expert 204 and suggestion new experiments 206 to approach the target parameter values. This may require a lot of time and slow down development time of chemical products. Further, chemicals and electricity are required for conducting experiments. In turn, capacities for conducting other experiments are blocked towards other experiments. Hence, it is desired to reduce the number of experiments conducted per development of a chemical product. This comes with the benefit of saving materials, electricity and time while accelerating material research by freeing up capacities for conducting further experiments. Fol lowingly , reducing the number of experiments allows for a more than linear increase in speed of developing new materials. A difference between a traditional chemical research as shown in FIG. 2A can be seen in FIG. 2B and the following Figures.

[0059] FIG. 2B illustrates an embodiment of determining one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0060] FIG. 2B illustrates a workflow that allows to reduce the number of experiments significantly. In comparison to the traditional workflow of FIG. 2A, the presented workflow in FIG. 2B allows to reduce the number of experiments to a third or a quarter of the experiments required by the traditional workflow while arriving at better results, e.g. better maximization of a target parameter as it can be seen in FIG. 7.

[0061] To do so, historical experimental data 202 may be gathered and provided together with a target parameter range 218 to an extraction data-driven model 214. From the provided data, the extraction data-driven model 214 may extract sets of parameter values associated with at least one target parameter and at least one parameter other than the target parameter. These sets may be indicative of a relation between the target parameter and the at least one parameter other than the target parameter. In particular, the sets may be indicative of an influence of the at least one parameter other than the target parameter on the target parameter. Based on these sets of parameter values, an operation data-driven model 216 may be determined. The operation data-driven model 216 may be configured to relate the parameter values associated with the target parameter to the parameter values associated with the at least one parameter other than the target parameter. Hence, the suggestion model may be indicative and / or may comprise one or more mathematical relation(s) between the parameter values associated with the target parameter and the parameter values associated with the at least one parameter other than the target parameter. From the operation data-driven model 216, a suggestion for improving the parameter values associated with the target parameter towards the target parameter range may be obtained, e.g. by determining parameter values associated with the parameter other than the target parameter corresponding to a parameter value associated with the target parameter closer to and / or within the target parameter range according to the operation data-driven model 216. The parameter other than the target parameter may be a tuneable parameter. The tuneable parameter may be influence the one or more target chemical reaction(s). Adapting the tuneable parameter values may result in adapting the parameter values associated with the target parameter. The parameter values associated with the target parameter may be adapted indirectly by adapting one or more tuneable parameter values. By the operation data-driven model 216, a different configuration of the tuneable parameter may be suggested. A new experiment may be conducted according to the suggested tuneable parameter value. From this validation experimental data may be obtained. The validation experimental data may be indicative of parameter values associated with the target parameter in relation to the suggested tuneable parameter value. The operation data-driven model 216 may be updated based on the updated parameter values associated with the target parameter and the suggested tuneable parameter value. This may increase the accuracy for determining the parameter values associated with the target parameter in relation to tuneable parameter values. Hence, the operation data-driven model 216 may be improved over a few amount of iterations. The so-obtained operation data-driven model 216 allows to arrive closer to the target range than the traditional workflow as described in the context of FIG. 2A while requiring less resource and time intensive iterations. Further details with regard to the processing of the historical experimental data 202 and the target parameter range 218, in particular by the extraction data-driven model 214 and / or the operation data-driven model 216 may be described in the following Figures.

[0062] FIG. 3 illustrates an embodiment of a method for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s).

[0063] A target parameter range may be provided 302. The target parameter range may be indicative of one or more target parameter value(s) associated with a target parameter characterizing one or more target chemical reaction(s). The parameter values associated with the target parameter may be adapted by adapting one or more tuneable parameter value(s). Hence, it may be desired to determine one or more tuneable parameter value(s) related to one or more target parameter value(s). The target parameter range may be provided via an interface such as a user interface. This may allow for tuning the parameters of a target chemical reaction to arrive at target parameter values. For example, it may be desired to minimize the amount of byproduct produced in a target chemical reaction. In the example, increasing the temperature may increase the conversion rate for obtaining the target chemical product over the byproduct until a threshold temperature may be reached. Meanwhile, increasing the temperature above the threshold temperature may increase the conversion rate of another chemical product than the target chemical product and the byproduct. Hence, a balance between increasing the temperature for increasing the conversion rate associated with the target chemical product and decreasing the temperature for decreasing the conversion rate associated with the other chemical product, i.e. finding the treshold temperature, may be desired, ie maximizing the conversion rate associated with obtaining the target chemical product. The target parameter range may specify one or more target parameter values for obtaining the target chemical product via one or more target chemical reaction(s) to allow for efficient production and / or processing of the target chemical product.

[0064] Experimental data indicative of a plurality of parameter values associated with a plurality parameters characterizing a plurality of chemical reactions may be provided 304. At least a part of the plurality of parameters may be tuneable parameters and / or may have an influence on the target parameter. In particular, adapting at least a part of the plurality of parameters may result in adapting the parameter values associated with the target parameter. The experimental data may be obtained by one or more sensor(s) associated with the plurality of chemical reactions. The experimental data may be provided by a database, i.e., a database may store the experimental data. In the above-described example, the experimental data may be indicative of one or more temperature curve(s) in relation to a concentration of the target chemical product at a predefined point in time. The experimental data may comprise numerical data, in particular tabular data, string data and / or image data, in particular image data indicative of a plot of one or more tuneable parameter values against the property values associated with the target parameter. The experimental data may be indicative of a plurality of parameter values of parameters characterizing a plurality of chemical reactions. The experimental data may be indicative of one or more tuneable parameters characterizing the target chemical reaction.

[0065] An applicable parameter range may be received and / or provided 306. The applicable parameter range may be received and / or provided via an interface such as an user interface. The applicable parameter range may be provided based on an inventory of a target chemical production facility for obtaining the target chemical product and / or one or more setting(s) related to the target chemical production facility. This allows to set limitations with regard to available means for obtaining the target chemical product. For example, where machinery may be operated within a predefined pressure range, obtaining the chemical product outside of the predefined pressure range may be a safety risk and / or may be impossible. Followingly, setting limitations allows to focus on feasible chemical reactions. Furthermore, this limits the search space for tuneable parameter values and in turn, improves the accuracy of the operation data-driven model. Thereby, less iterations for adapting the operation data-driven model are required and hence, material and time can be saved. Additionally or alternatively, the applicable parameter range may be determined by an extraction data-driven model and / or based on the experimental data. The experimental data may be indicative of a plurality of parameter values of parameters characterizing a plurality of chemical reactions including the one or more target chemical reaction(s). The experimental data may be indicative of a subset of parameter values characterizing a plurality of chemical reactions. The subset of parameter values may correspond to the applicable parameter range. Hence, by evaluating the experimental data, the applicable parameter range may be determined. In an embodiment, the experimental data may be provided to the extraction data-driven model for extracting the applicable parameter range. The extraction data-driven model may be configured to extract and / or suitable for extracting parameter values from provided input data, i.e. the experimental data. Examples of extraction data-driven models may be described in further detail the context of FIG. 8. The extraction data- driven model may be provided with the experimental data. Further, the extraction data-driven model may be provided with range extraction task instructions. The range extraction task instructions may trigger the extraction data-driven model to extract the applicable parameter range from the experimental data. The range extraction task instruction may be indicative of instructions to the extraction data-driven model. The extraction data-driven model may be configured to follow task instructions provided to the extraction data-driven model. For example, the extraction data-driven model may be a large language model and / or the task instructions may comprise string data indicative of the instructions to be carried out by the extraction data-driven model. By determining the applicable parameter range from the experimental data, a meaningful parameter space can be defined according to knowledge gained from previous experiments or experiments performed by other entities. Further, this enables to convey insights into chemical phenomena from historical experiments to improve efficiency of future experiments.

[0066] At least a part of the experimental data may be selected according to the applicable parameter range and / or the one or more target chemical reaction(s) 308. The experimental data may be associated at least in parts with parameters outside of the applicable parameter range. At least the part of the experimental data may be selected based on the inventory of a target chemical production facility for obtaining the target chemical product and / or one or more setting(s) related to the target chemical production facility. This allows to set limitations with regard to available means for obtaining the target chemical product. For example, where machinery may be operated within a predefined pressure range, obtaining the chemical product outside of the predefined pressure range may be a safety risk and / or may be impossible. Followingly, setting limitations allows to focus on feasible chemical reactions. Furthermore, this limits the search space for tuneable parameter values and in turn, improves the accuracy of the operation data-driven model 216. Thereby, less iterations for adapting the operation data-driven model are required and hence, material and time can be saved. Further, experimental data may apply to a subset of chemical reactions including the target chemical reactions. Hence, selecting at least the part of the experimental data according to the one or more chemical reaction(s) allows to use the applicable part of the experimental data and exclude non-meaningful experimental data from data analysis. In an embodiment, selecting at least the part of the experimental data may comprise retrieving at least the part of the experimental data by providing a query based on the applicable parameter range and / or the one or more target chemical reaction(s), in particular to the database comprising the experimental data. Additionally or alternatively, selecting at least the part of the experimental data may trigger at least a part of available sensor(s) to record experimental data associated with the one or more target chemical reaction(s).

[0067] Additionally or alternatively, selecting at least the part of the experimental data may comprise determining if parameter values associated with the experimental data may be within the applicable parameter range and selecting the part of the experimental data associated with the parameter values within the applicable parameter range. Additionally or alternatively, selecting the part of the experimental data may comprise determining a numerical representation of the experimental data and determining a distance score indicative of a distance between the numerical representation of the target parameter range and the numerical representation of the experimental data. The numerical representation(s) may be obtained by one or more embedding layer(s) as described in the context of FIG. 8. In an example, the target parameter range may comprise string data, numerical data such as tabular data, image data or the like. The numerical representation of the target parameter range and / or the experimental data may allow to select the part of the experimental data according applicable parameter range in a computationally low-cost manner even if the target parameter range may not allow for direct comparison with the experimental data, e.g. because of a different modality of the target parameter range than the experimental data.

[0068] Proposed experimental data may be determined according to the selected part of the experimental data 310. The proposed experimental data may be determined from at least two parameter values associated with at least two different parameters. The at least two different parameters may be related to each other. Preferably, at least one of the two different parameters may be the target parameter and the at least one other parameter may be the parameter other than the target parameter, in particular the at least one tuneable parameter. Determining the proposed experimental data may comprise determining at least two proposed parameter values associated with the at least two different parameters from the at least two parameter values associated with the selected part of the experimental data. A cumulative distribution function may be determined from the selected part of the experimental data. The cumulative distribution function may be indicative of a distribution of parameter values, in particular a probability of an appearance of parameter values associated with the at least two different parameters. The cumulative distribution function may be divided into a number of parts equal to a number of corresponding parameter values and / or a number of sets of parameter values. For example, where two parameter values may be associated with two different parameters, one further set of two parameter values associated the two different parameters may be determined. Where three sets of parameter values, i.e. three parameter values associated with a first parameter and three parameter values associated with a second parameter different from the first parameter, may be used for determining the proposed experimental data, the cumulative distribution function may be divided into three parts. A set of parameter values, i.e., at least two parameter values, may be determined per part of the cumulative distribution function. This may be known as latin hypercube sampling. Additionally or alternatively, the proposed experimental data may be determined by monte carlo sampling. In general, a plurality of sampling methods may be available to obtain proposed experimental data from measured experimental data, i.e. , the selected part of the experimental data.

[0069] Additionally or alternatively, the proposed experimental data may be determined by providing an extraction task instruction for extracting proposed experimental data including the selected part of the experimental data to a data-driven model. The data-driven model may be configured to follow task instructions, in particular may be the extraction data-driven model. The data-driven model, in particular the extraction data-driven model may be a model as described in the context of FIG. 8. The data-driven model may provide the proposed experimental data in response to receiving the extraction task instruction. The data-driven model may be suitable for and / or configured to extract proposed experimental data from the selected part of the experimental data. In particular, the selected part of the experimental data may be associated with and / or indicative of one or more relations between two or more parameters. Hence, the data-driven model may obtain the proposed experimental data from a relation between two or more parameters, in particular upon receiving the extraction task instruction. The extraction task instruction may further comprise the applicable parameter range. This may guide the data-driven model to generate the proposed experimental data within the applicable parameter range. The relation between the two or more parameters may be suitable for determining at least one parameter value associated with at least one of the two or more parameters from another parameter value associated with at least another one of the two or more parameters. Further, the relation between the at least two parameters may be obtained by retrieving the relation between the least two parameters from a data storage such as a database. The proposed experimental data may be obtained by applying the extracted and / or retrieved relation.

[0070] By doing so, more sample points can be used for building a operation data-driven model. This improves the performance, in particular the precision and the accuracy, of said operation data-driven model. Overall, this result in reducing the number of experiments to be carried out, hence reducing resource invest for obtaining a target chemical product.

[0071] The proposed experimental data and / or the selected part of the historical experimental data may be provided to an extraction data-driven model for extracting at least one set of parameter values associated with at least one target parameter and at least one parameter other than the target parameter 312. The extraction data- driven model may be suitable for extracting and / or configured to extract at least one set of parameter values associated with at least one target parameter and at least one parameter other than the target parameter. The extraction data-driven model may be configured to follow task instructions. The task instruction may trigger the extraction data-driven model to extract the at least one set of parameter values. The task instruction may include the selected part of the experimental data and / or the proposed experimental data. In particular, the at least one target parameter and the at least one parameter other than the target parameter may be related, i.e. adapting the at least one parameter other than the target parameter may result in adapting the target parameter. This may allow for determining parameter values of the target parameter in relation to a tuneable parameter associated with a target chemical reaction. Usually, a plurality of tuneable parameters may be available. The experimental data may be indicative of a plurality of relations, e.g. by providing a plurality of sets of parameter values. The extraction data-driven model may be suitable for selecting the relevant tuneable parameters from the data input by determining parameter values of the target parameter according to the data input in relation to at least one tuneable parameter. In other words, the extraction data-driven model can extract not explicitly disclosed parameter values by obtaining relations between parameters indicated by experimental data. This allows to extract more information from already available experimental data without the need for generating new experimental data. Hence, the extraction data-driven model may interpolate and / or extrapolate from measured experimental data to determine parameter values of the target parameter in relation to a tuneable parameter. By using more information from already available sources, the number of newly conducted experiments can be lowered significantly. Ultimately, this allows to reduce material, energy, water and time used for experimentation in order to obtain target chemical products.

[0072] An operation data-driven model may be determined based on the at least one extracted set of parameter values 314. Determining the operation data-driven model may comprise configuring the operation data-driven model to map parameter values associated with parameters other than the target parameter to parameter values associated with the target parameter. The operation data-driven model may be a data-driven model, i.e., may obtain a relation between the target parameter and at least one parameter other than the target parameter from the extracted set of parameter values. A plurality of different data-driven models may be available to relate parameters. For example, the operation data-driven model may include a neural network, a regression model such as linear regression, use bayesian optimization or the like. Determining the operation data-driven model may comprise updating model parameters associated with the operation data-driven model to decrease a distance between the parameter values provided by the operation data-driven model and the parameter values indicated by at least a part of the extracted set of parameter values, i.e. the parameter values associated with the target parameter. In the case of a neural network, determining the operation data- driven model may include initializing model parameter values indicative of a structure, i.e. a number of nodes and one or more connection(s) between the nodes, determining a parameter value associated with the target parameter by the operation data-driven model from at least one parameter value associated with the parameter other than the target parameter, determining a deviation between the determined parameter value and the extracted parameter value associated with the target parameter. Said parameter values may be related to the at least one parameter value associated with the parameter other than the target parameter. The deviation may be determined by a cost function associated with the operation data-driven model. Depending on the deviation, a measure for updating the model parameters may be determined via backpropagation of the deviation. The measure for updating the model parameters may be determined per model parameter by adjusting a cost function associated with the operation data-driven model via gradient descent. In the case of bayesian optimization, a plurality of functions including at least a part of the experimental data may be determined. The operation data-driven model may comprise a mean function representing a mean between the plurality of functions determined. The plurality of functions may be determined within a predefined range from the mean function, i.e. a standard deviation. In the case of linear regression, model parameters of a linear predictor function may be updated to decrease a deviation between the determined parameter value and the extracted parameter value associated with the target parameter.

[0073] The operation data-driven model may be provided and / or deployed 316. Providing the operation data-driven model may include for example to provided the operation data-driven model from a model providing engine 514 to a suggestion providing engine 516. This may allow for use of the operation data-driven model. Using the operation data-driven model may comprise updating the operation data-driven model by updating one or more model parameter(s) of the operation data-driven model. Using, adapting and / or deploying the operation data-driven model may be described further in FIG. 4. FIG. 4 may continue the method as described in the context of FIG. 3.

[0074] In an embodiment, the experimental data may comprise text data, numerical data, in particular tabular data, image data or a combination thereof. The data-driven model used for processing the experimental data may be associated with different embeddings schemas as described in the context of FIG. 8.

[0075] FIG. 4 illustrates an embodiment of a method for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0076] The operation data-driven model may be obtained as described in the context of FIG. 3. FIG. 4 may show deploying, using and / or updating the operation data-driven model.

[0077] Producing target chemical products may require developing production conditions, i.e. chemical reaction parameters, for producing the target chemical product associated with a target parameter value. For example, known conversion rates may be too slow for large-scale chemical production, chemical product may not be suited for a target application because the parameter value of the target parameter may be outside of a range indicated by a target parameter range, i.e. the chemical product may not be associated with a required performance, or share of the target chemical product produced by a chemical reaction may be low and consequently a large amount of byproducts, i.e. waste, may be produced. All of these examples require adaption of parameters of the production process such as composition, catalyst, reaction conditions or the like to arrive at the target chemical product associated with the target parameter value of the target parameter. The operation data-driven model may be used to approach the target parameter value. Nevertheless, confirming experimentation may be necessary to confirm the operation data-driven model or to identify potential for adaption. As described in the context of FIG. 2B, the operation data-driven model may be updated upon newly generated experimental data. For conducting further experiments, i.e. providing further experimental data, to confirm and / or adapt the operation data-driven model, suggestion parameter, it is beneficial to use an educated guess provided by the operation data-driven model. This allows to arrive faster with less experiments at a better operation data-driven model as currently available as it can be seen in FIG. 7.

[0078] One or more suggestion parameter value(s) related to at least one target parameter may be determined by the operation data-driven model 402. The suggestion parameter value(s) may be related to the at least one target parameter value by the operation data-driven model. Hence, the operation data-driven model may determine the suggestion parameter value from the at least one target parameter. In layman's words, the operation data- driven model provides tuneable parameters of a target chemical reaction for the parameter value of the target parameter to be within the target parameter range.

[0079] Upon determining the suggestion parameter value(s), an experiment may be conducted for confirming the suggestion from the operation data-driven model. The experiment may be conducted according to the suggestion parameter value. While conducting the experiment, experimental data indicative of at least one parameter value associated with the at least one target parameter may be obtained. The obtained experimental data may be verification experimental data. The verification experimental data may allow for verifying the suggestion by the operation data-driven model. The verification experimental data may be provided 404.

[0080] In an embodiment, the operation data-driven model may comprise the extraction data-driven model and a suggestion sub model. The suggestion sub model may be determined as the operation data-driven model as described in the context of 314. The operation data-driven model may be a combination of two or more data- driven models. Determining the suggested parameter values by the operation data-driven model may include determining the suggested parameter values by two or more data-driven models, in particular the extraction data-driven model and the suggestion sub model and relating the determined suggested parameter values via two or more weighting factors associated with the two or more models. Additionally or alternatively, the operation data-driven model may comprise a combined function. The combined function may be obtained by relating two or more function(s) associated with the two or more data-driven models via two or more weighting factors associated with the two or more data-driven models. The combination of the two or more data-driven models may be obtained by transfer learning. Hence the operation data-driven model may be obtained by combining the weights associated with the two or more data-driven models.

[0081] It may be determined if the operation data-driven model may be updated according to the verification experimental data 406. This may include comparing the parameter value associated with the target parameter determined by the operation data-driven model with the parameter value associated with the target parameter provided with and / or indicated by the verification experimental data. Optionally, the parameter value associated with the target parameter may be determined from the verification experimental data. This parameter value may be the verification parameter value. If the parameter value associated with the target parameter determined by the operation data-driven model correspond to the verification parameter value, the operation data-driven model may be confirmed and / or an indication of a confirmation of the operation data- driven model by the verification experimental data may be provided. Otherwise, the operation data-driven model may be updated based on the verification experimental data 410. Updating the operation data-driven model based on the verification experimental data may include the steps of determining the operation data- driven model by starting from the operation data-driven model and updating the model parameters of the operation data-driven model according to the verification experimental data. Hence, updating the operation data-driven model may not include initializing the operation data-driven model. Updating the operation data- driven model may initialize and / or trigger repeating 402 to 406 until the parameter value associated with the target parameter determined by the operation data-driven model correspond to the verification parameter value. Parameter values corresponding to each other may comprise refer to the parameter value associated with the target parameter determined by the operation data-driven model being within a predefined distance from the verification parameter value. Once, the parameter value associated with the target parameter determined by the operation data-driven model corresponds to the verification parameter value, the operation data-driven model may be provided and / or deployed analogous to 316.

[0082] By following this approach, the operation data-driven model is improved iteratively. This allows to assess whether further experiments are needed per iteration cycle. Hence, only a number of experiments required to fulfill the set targets is conducted. Consequently, resources are applied efficiently and waste of water, material, energy and time can be circumvented.

[0083] FIG. 5 illustrates an operating system of a chemical production entity 118 for obtaining a target chemical product by one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing one or more target chemical reaction(s).

[0084] The operating system of a chemical production entity 118 may be connected to a control engine of a chemical production entity 518. The operating system of a chemical production entity 118 may provide parameter values for controlling the one or more target chemical reaction(s) to the control engine of a chemical production entity 518. The control engine of a chemical production entity 518 may be configured to control the chemical product entity according to the provided parameter values. The operating system of a chemical production entity 118 may be configured to determine the parameter values, e.g. by performing the method according to FIG. 4 and FIG. 3. For this purpose, the operating system of a chemical production entity 118 may comprise an intake interface 504, an experimental data proposing engine 508, a selection engine 506, an extraction engine 510, a suggestion engine 512 and / or an output interface 502.

[0085] The intake interface 504 may be configured to receive a target property range. The target property range may indicate at least one threshold target parameter value. The target property range may be defined according to the at least one threshold target parameter value. Hence, the intake interface 504 may be configured to receive the threshold target parameter value. The threshold target parameter value may be provided to the experimental data proposing engine 508 and / or the selection engine 506. In an embodiment, only proposed experimental data, only a selected part of the experimental data and / or a combination of the proposed experimental data and the selected part of the experimental data may be used for determining the operation data-driven model. The experimental data may be stored by a database 520. The database 520 may be configured to provide the experimental data to the experimental data proposing engine 508 and / or the selection engine 506, in particular upon providing a request for retrieving the experimental data to the database 520. The selection engine 506 may be configured to selected at least a part of the experimental data as described in the context of FIG. 3. The experimental data proposing engine 508 may be configured to determine the proposed experimental data as described in the context of FIG. 3. The proposed experimental data and the selected part of the experimental data may be provided to the extraction engine 510. The extraction engine 510 may be configured to extract at least one set of parameter values associated with at least one target parameter and at least one parameter other than the target parameter as described in the context of FIG. 3. The extracted parameter values may be provided to the suggestion engine 512 for determining the operation data-driven model and providing suggestion parameter values by the operation data-driven model. The suggestion engine 512 may comprise a model providing engine 514 and a suggestion providing engine 516. The model providing engine 514 may be configured to determine the operation data- driven model as described in the context of FIG. 3. The suggestion providing engine 516 may be configured to determine the suggestion parameter value(s) as described in the context of FIG. 4. The model providing engine 514 may provide the operation data-driven model 216 to the suggestion providing engine 516. The suggestion parameter values may be provided to the output interface 502. The output interface 502 may provide the suggestion parameter values to the control engine of a chemical production entity 518 for controlling the production of the target chemical product.

[0086] FIG. 6 illustrates an embodiment of a user interface for obtaining a target chemical product.

[0087] The user interface 602 may allow for determining instructions for obtaining the target chemical product associated with at least one target parameter value of a target parameter. Hence, the user interface 602 may comprise and / or may correspond the intake interface 504 and / or the output interface 502 as described in the context of FIG. 5.

[0088] The user interface 602 may allow for entering the applicable parameter range, the target parameter range and / or an indication of the one or more target chemical reaction(s). In an embodiment, the one or more target chemical reaction(s) may be indicated by the experimental data provided, e.g. experimental data characterizing the one or more target chemical reactions. Optionally, the indication of the one or more target chemical reaction(s) may be provided. The indication of the one or more target chemical reaction(s) may comprise an indication of the one or more educt(s) and / or product(s) and / or one or more reaction condition(s) associated with the one or more target chemical reaction(s). In an embodiment, the experimental data may be provided via the user interface, e.g. a user may upload a selection of experimental data.

[0089] FIG. 7 illustrates evolution of a parameter values of a target parameter in relation to a number of iterations including conducting experiments.

[0090] In the example, the target parameter corresponds to a discharge capacity of a battery material. The tuneable parameter may be a composition of the battery material and / or reaction conditions for combining the components of the battery material. An iteration may include analyzing experimental data, providing suggested parameter values for controlling the one or more target chemical reaction(s), conducting one or more experiment(s) for verifying the suggested parameter values and verifying if the suggested parameter value resulted in at least one target parameter value within the target parameter range. Upon determining that the suggested parameter value did not result in the at least one target parameter value within the target parameter range, at least another iteration may be necessary to arrive at the at least one target parameter value. From FIG. 7 it can be concluded that random guessing results in less performant battery materials than compared to the methods as described herein, in particular in the context of FIG. 2B, FIG. 3 and FIG. 4.

[0091] FIG. 8 illustrates an embodiment of a data-driven model, i.e. a transformer encoder comprising an encoder input 888, one or more encoder block(s) 886 and an encoder output 876 and / or a transformer decoder comprising a decoder input 894, one or more decoder block(s) 890 and a decoder output 892 and / or a transformer encoder-decoder.

[0092] The transformer encoder comprises an encoder input 878, one or more encoder blocks 874, 814 and an encoder output. A plurality of transformer encoder architectures are available in the art such as the bidirectional encoder representations from transformers (BERT). The input data may be received at the encoder input 878. The input data may comprise at least one of text data, numerical data, tabular data, image data or the like. Where the input data may comprise one of text data, numerical data, tabular data, image data or the like, input embedding of a type corresponding to the type of input data may be applied. Hence, the input embedding may be configured to map text data, numerical data, tabular data, image data or a combination thereof to a numerical representation of the input data. An example of input embedding associated with text data may be a continuous bag-of-words-model (CBOW). Additionally or alternatively, Word2Vec may be used for representing input data. Upon receiving the input data, the input data may be tokenized via a vocabulary associated with a preselection of elements of expected input data. Applying the input embedding may comprise mapping the input data, in particular the two or more elements of the input data to a numerical represenation of the input data, preferably of a predefined size e.g. via padding. Further, the encoder input 878 may apply positional encoding 804. For example, the positional factor pposmay be obtained based on the following equation: pos ppos(2i + l) = cos ( - -) lOOOOd- where pos may refer to the position of the element within the sequence, / may refer to the dimension associated with the input embedding and d may refer to the dimension of the data-driven model. Alternatively, the positional encoding may be based on rotary positional embeddings (RoPE). The embedded input data may be processed by the encoder block. The embedded input data may be provided to the layer normalization 808 by a residual connection. Multi-head self-attention 806 may be applied to the embedded input data. The embedded input data may serve as query Q, key K and value V with respect to the self-attention operation. For improving the efficiency, multiple heads are used to apply the filter according to the following equation: head i = Attention^ with parameter matrices where i may refer to the number of heads, dv, dKand dQmay refer to the dimensions of the value, key and query.

[0093] The result of the two or more head may be concatenated according to the following equation:

[0094] MultiHead Q, K, P) = Concat(head 1, . . . , headh)W° where Woe ]R> xd and h may refer to the number of heads.

[0095] This may result in a context tensor. After the multi-head self-attention 806 layer normalization 808 may be applied based on the context tensor and / or the embedded input data from the residual connection. The so- obtained tensor may be passed to a feed-forward layer 810 again followed by layer normalization 812 based on the residual connection to the context tensor and / or the output of the feed-forward layer 810.

[0096] The encoder output 876 may comprise a linear layer 834 and a softmax layer 836. The encoder output may be configured to map the concatenated numerical representation of the two or more elements and the relation between the two or more elements to a numerical representation of the output data. Additionally or alternatively, a decoding model may be used to map the concatenated numerical representation of the two or more elements and the relation between the two or more elements to a numerical representation of the output data. The decoding model may be trained to relate a numerical represenation of data of a type according to the input data.

[0097] The transformer decoder comprises a decoder input 884, one or more decoder blocks 880, 828 and a decoder output 892. A plurality of transformer decoder architectures are available in the art such as the generative pretrained transformers (GPT). In contrast to the transformer encoder, the transformer decoder may perform masked multi-head self-attention 820 by additionally masking a part of the embedded input data associated with elements later in the sequence than the element to be generated. Additionally or alternatively, the part of the input data associated with elements later in the sequence than the element to be generated may not be received and / or transformed into the embedded input data.

[0098] The transformer encoder-decoder may comprise a combination of the transformer encoder and transformer decoder wherein the context tensor obtained from the encoder block may be used for the multi-head selfattention 864 operation in at least one decoder block 890.

[0099] Input data to the data-driven model may comprise image data. The encoder input and / or decoder input may comprise one or more linear projection layer(s) for a linear projection of a sequence of two or more partial images. This may result in changing the dimension of the one or more received images. Furthermore, positional embedding may be applied to the sequence, preferably by passing the sequence of one or more images and / or partial images through the one or more linear projection layer(s). Additionally or alternatively, the input data may comprise tabular data. Input embeddings for tabular input data may comprise a token embedding, a positional embedding, a column embedding, a row embedding or a combination thereof.

[0100] In an example, the data-driven model may comprise a Mamba block. A Mamba architecture may enhance inference speed in relation to a transformer based model. Mamba block may be based on a selective space state sequence model. A selective state space layer may be a linear recurrent network that selectively process data based on the input token, which may allow to focus on relevant data and discard irrelevant data. For instance in each step a separate weight vector may be determined based on the respective input token. The determined weight vector may then be used in a selective scan. A selective state space layer may be used in a convolutional mode e.g. for parallelizable training and a recurrent mode for near-constant time generation of output data. A state space operation may be based on solving the state and output equations, wherein a state equation may describe how a state changes based on how the input influences the state and an output equation may describe how the state is translated to the output. Further how the input influences the output may be represented by a learnable linear transformation used in a learnable skip connection. An example of the architecture of a mamba block may be found in "Mamba: Linear-Time Sequence Modeling with Selective State Spaces” by Albert Gu and Tri Dao arXiv:2312.00752v2 [cs.LG] 31 May 2024, , which is incorporated herein by reference.

[0101] In an example, any one of the data-driven models may further comrpise a mixture of experts block. The mixture of experts block may be decoder blocks wherein the feed-forward layer may be exchanged for a gating network and a number of parallel feed-forward layers, wherein the gating network may switch between the feed-forward layers depending on the input. This may allow leveraging advantages of the different architectures. The data-driven model may generate one token per timestep upon processing the input data and optionally all output tokens already produced. The training data set may comprise a plurality of sequences comprising a plurality of elements. During the training of the data-driven model, sequences of the training data set may be provided to the data-driven model and one or more elements may be generated based on the sequences of the training data set one by another. The elements generated based on the sequences may follow the elements of the parts of sequences the data-driven model may have been provided with. The generated one or more elements may be compared to the one or more elements following the at least a part of the sequences provided to the data-driven model in order to adapt the parameters of the data-driven model depending on a deviation of the predefined token and the generated token.

[0102] The present disclosure has been described in conjunction with preferred embodiments and examples as well. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed subject-matter, from the studies of the drawings, this disclosure and the claims. Notably, in particular, the any steps presented can be performed in any order, i.e. the present disclosure is not limited to a specific order of these steps. Moreover, it is also not required that the different steps are performed at a certain place or at one node of a distributed system, i.e. each of the steps may be performed at different nodes using different equipment / data processing.

[0103] As used herein ..determining" also includes ..initiating or causing to determine", "generating" also includes ..initiating and / or causing to generate" and "providing” also includes "initiating or causing to determine, generate, select, send and / or receive”. "Initiating or causing to perform an action” includes any processing signal that triggers a computing node or device to perform the respective action.

[0104] In the claims as well as in the description the word "comprising” or "including” or similar wording does not exclude other elements or steps and shall not be construed limiting to the elements or steps lined out. The indefinite article "a” or "an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation or further elements may be included.

[0105] Providing in the scope of this disclosure may include any interface configured to provide data. This may include an application programming interface, a human-machine interface such as a display and / or a software module interface. Providing may include communication of data or submission of data to the interface, in particular display to a user or use of the data by the receiving entity.

[0106] Any disclosure and embodiments described herein relate to the methods, the systems, devices, the computer program element lined out above and vice versa. Advantageously, the benefits provided by any of the embodiments and examples equally apply to all other embodiments and examples and vice versa.

Claims

CLAIMS1. A method for controlling one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s), the method comprising: providing a target parameter range indicative of one or more target parameter value(s) associated with a target parameter characterizing the one or more target chemical reaction(s) determining one or more suggestion parameter value(s) associated with a tuneable parameter by an operation data-driven model from the one or more target parameter value(s), wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, and wherein the operation, data-driven model is determined by receiving experimental data comprising a plurality of parameter values associated with a plurality of parameters characterizing the one or more target chemical reaction(s), wherein the plurality of parameters comprises at least one target parameter and at least one tuneable parameter, wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, and providing, to an extraction data-driven model, model determining task instructions including the experimental data for providing the operation data-driven model, wherein the extraction data- driven model is configured to follow task instructions, and wherein the operation data-driven model is configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data, providing the one or more suggestion parameter value(s), in particular for controlling the one or more target chemical reaction(s) according to the one or more suggestion parameter value(s).

2. The method of claim 1 , further comprising providing verification experimental data indicative of at least one verification parameter value associated with the at least one target parameter, wherein the verification experimental data is obtained during the one or more target chemical reaction(s) controlled according to the one or more suggestion parameter value(s) and providing a trigger for adapting the operation data-driven model in response to determining that the operation data-driven model is to be updated upon the verification experimental data.

3. The method of any one of claims 1 or 2, wherein the operation data-driven model comprises a combination of two or more data-driven model(s) configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data.

4. A method for determining an operation data-driven model configured to determine parameter values for controlling one or more target chemical reaction(s), the method comprising: receiving experimental data comprising a plurality of parameter values associated with a plurality of parameters characterizing the one or more target chemical reaction(s), wherein the plurality of parameters comprises at least one target parameter and at least one tuneable parameter, wherein adapting the parameter values associated with the tuneable parameter influences the parameter values associated with the target parameter, providing model determining task instructions including the experimental data for providing the operation data-driven model to an extraction data-driven model, wherein the extraction data- driven model is configured to follow task instructions, and wherein the operation data-driven model is configured to relate parameter values associated with the at least one tuneable parameter and the at least one target parameter based on the experimental data, providing the operation data-driven model.

5. The method of any one of claims 1-4, wherein providing model determining task instructions to the extraction data-driven model comprises: providing extraction task instructions including the experimental data to an extraction data-driven model for extracting at least one set of parameter values associated with the at least one target parameter and the at least one tuneable parameter from the experimental data, wherein the extraction data-driven model is configured to follow task instructions, and training the operation data-driven model to relate parameter values associated with the at least one tuneable parameter to the at least one target parameter according to the at least one extracted set of parameter values.

6. The method of any one of claims 1-5, wherein determining task instructions to the extraction data- driven model comprises: providing model determining task instructions including the experimental data for generating the operation data-driven model to an extraction data-driven model, and wherein the operation data-driven model is provided by the extraction data-driven model.

7. The method of any one of claims 1-6, wherein the extraction data-driven model is trained based on historical extraction task instructions and corresponding extracted sets of parameter values associated with two or more parameters.

8. The method of any one of claims 1-7, further comprising determining proposed experimental data according to a distribution of the parameter values associated with the at least one target parameter and the at least one tuneable parameter within the experimental data, and / or determining the proposed experimental data according to a relation between the parameter values associated with the at least one target parameter and theparameter values associated with the at least one tuneable parameter, wherein the proposed experimental data comprises at least one proposed set of parameter values associated with at least one target parameter and at least one tuneable parameter, and wherein the extraction task instructions further include the proposed experimental data.

9. The method of any one of claims 1-8, wherein the experimental data comprises proposed experimental data including at least one proposed set of parameter values associated with at least one target parameter and at least one tuneable parameter, wherein the proposed experimental data are determined according to a distribution of the parameter values associated with the at least one target parameter and the at least one tuneable parameter within historical experimental data, and / or determining the proposed experimental data according to a relation between the parameter values associated with the at least one target parameter and the parameter values associated with the at least one tuneable parameter.

10. The method of any one of claims 1-9, further comprising providing an applicable parameter range based on an inventory of a target chemical production facility for obtaining the target chemical product and / or one or more setting(s) related to the target chemical production facility and selecting at least a part of the experimental data according to the applicable parameter range, and wherein the experimental data provided to the extraction data-driven model corresponds to the selected part of the experimental data.

11. The method of any one of claims 1-10, further comprising determining an applicable parameter range from the experimental data by the extraction data-driven model, wherein the applicable parameter range is indicative of a range of tuneable parameter values associated with at least one tuneable parameter, and wherein the experimental data provided to the extraction data-driven model corresponds to the selected part of the experimental data.

12. The method of any one of claims 1-11 , wherein the experimental data is indicative of the one or more target chemical reaction(s) and / or further providing the indication of the one or more target chemical reaction(s), and wherein a part of the experimental data is selected according to the indication of the one or more target chemical reaction(s), and wherein the experimental data provided to the extraction data-driven model corresponds to the selected part of the experimental data.

13. Use of an extraction data-driven model for extracting at least one set of parameter values associated with at least one target parameter and at least one tuneable parameter for training an operation data- driven model according to any one of the claims 4-12 based on the at least one set of parameter values.

14. Use of an operation data-driven model trained according to any one of the claims 4-12 for determining parameter values for controlling one or more target chemical reaction(s).

15. An apparatus for controlling one or more target chemical reaction(s) according to one or more parameter value(s) associated with one or more parameter(s) characterizing the one or more target chemical reaction(s), the apparatus comprising: a processor; and a memory storing instructions that, when executed by the processor, configure the apparatus to perform the steps of any one of the methods of any one of claims 1 to 12.

Images